Wellsware products are appliance storage that provide the best performance required by customers.
It can support up to 200Gbs (12.5GB/s) of bandwidth to support high performance, providing the customer's desired Performance can be scaled up. Wellsware, Performance Scale-up Storage (PerSS), helps customers by dramatically shortening the time, we are helping to create more business opportunities for our customers.
Satisfying performance and capacity at the same time
A 2-hour movie has approximately 20 hours of data source content.
This data source requires more than 50 to 100 TB of data space.
If completed data sources saved separately and several layers of work need to be done simultaneously, superior performance and storage space are required.
High-definition video processing environment
Amount of data required depending on video format
Video Format
Data Amount
HD
1,920×1,080, YUV4:2:0, 8 bits, 30fps
746Mbps
4K UHD
3,840x2,160, YUV4:2:0, 8bits, 30fps
3Gbps (4 times that of HD)
3,840x2,160, YUV4:4:4, 12bits, 60fps
18Gbps (24 times that of HD)
8K UHD
7,680x4,320, YUV4:2:0, 8bits, 30fps
12Gbps (16 times that of HD)
7,680x4,320, YUV4:4:4, 12bits, 60fps
72Gbps (96 times that of HD)
Problems with existing storage in video work
Existing products have small bandwidth/performance allocated to a single volume, so when higher performance is required, multiple physical volumes can be logically clustered or
Host connection ports are used by clustering, but in the case of these clustered volumes, data input/output throughput is not stable, resulting in performance fluctuations.
In high-definition video, screen interruption occurs, making it difficult to check the quality of the video.
Existing product
VS
Wellsware development product
In recent Multi-Cam editing environments, the number of video layers that need to be processed simultaneously has increased
AI Environment
Data in the AI environment may randomly request small-sized files, and depending on the workload, large-sized files such as images or videos may be requested sequentially.
Building a deep learning model on request is a process of executing various I/O patterns on storage.
Wellsware supports NVIDIA GPU direct access utilizing the RDMA protocol. It is also highly compatible with NVIDIA Mellanox switches.
NVMe-based storage for performance purposes is effective for processing up to hundreds of TB of data, but the major obstacle is the cost burden of storing and processing more than PB.
In order to build a true AI environment, not only large-scale data storage but also high-performance data processing is essential.
There is no “one-size-fits-all” storage solution for AI-powered apps.
Performance throughput is important for learning, latency is important for inference, and what will be the storage expectations as training data and inference data increase?
Clearly, as training/inference workloads grow, performance and capacity must also increase.
For the best AI performance, you should only consider scalable storage solutions that have the performance to keep your GPUs running without interruption.
After fully understanding the data set of the AI app, it is reasonable to select HDD only or hybrid NVMe/HDD method according to budget.
As complex workloads require ultra-fast processing of high-resolution simulations, very large data sets, and highly parallelized algorithms,
Compatibility with the NVIDIA ConnectX network must also be specifically considered.
Because RDMA technology can bypass the OS and dramatically reduce CPU resources, it has the advantage of being able to use the CPU's idle resources for other tasks, and can quickly transfer data from the story so that the GPU core can calculate quickly.
We refuse to collect email addresses without permission.
We refuse unauthorized collection of e-mail addresses posted on this website using e-mail collection programs or other technical devices, and please keep in mind that any violation of this will be subject to criminal punishment under the Information and Communications Network Act.
